Do you see an obvious performance problem there?

Yes, the function is definitely not super efficient.

Is it necessary to create a new KeySet and to convert all of the keys to a direct child just to find out if it is an array parent?

Nope, although I think you should put “just” under quotes here. Ideally checking for an array parent should be nothing more than searching for the meta key array. In reality storage plugins might not add that meta key at all though. As far as I remember YAJL is one of those plugins:

kdb mount config.json user/tests/yajl yajl
kdb set user/tests/yajl/array/#0 bla
#> Create a new key user/tests/yajl/array/#0 with string "bla"
kdb set user/tests/yajl/array/#1 blubb
#> Create a new key user/tests/yajl/array/#1 with string "blubb"
kdb lsmeta user/tests/yajl/array
#> 

. The commands above show a kdb session after I removed directoryvalue from infos/needs. If we do not assume that the meta key array is set correctly, then we have to check if every child of a key is an array key. That is more or less what isArrayParent does.

Shouldn't it be possible to do the check without modifying the keys?

Of course, but I guess that would mean more code and duplicated functionality. For now I modified isArrayParent locally, so it does not create a new KeySet anymore. The function still duplicates and modifies child keys though…

From a wild guess the duplication and modification of all keys causes the problem. Duplication of KeySets is quite fast (one memcopy and updating the ref counters).

A solution that requires the presence of the "array" metadata is imho the way to go, see also #182. It will also be needed by the high-level API. From a quick look only yajl and csvstorage seem to be the current offenders?

But even if we have the metadata, we nevertheless need to check if the structure and the metadata matches. So we need an efficient checking anyway.